Wind turbine tower hoisting screw loosening detection method based on feature difference distance threshold
Through the detection method based on the distance threshold of feature difference, image data and neural networks are used to identify the screw position and calculate the screw feature vector to achieve fast and accurate detection of loose screws for fan tower lifting, which solves the problem of low detection efficiency and accuracy in the existing technology The problem reduces the risk of wind motor collapse accidents.
Patent Information
- Application Number
- CN202410421443.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-09
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-04-09
AI Technical Summary
The existing methods for loose screws for fan tower lifting are low efficiency and accuracy, which leads to long-term patience and physical strength challenges, and is prone to errors, which increases the risk of wind turbine collapse accidents.
By using a detection method based on the characteristic difference distance threshold, by obtaining the image data of the fan tower lifting screws, using the Yolo neural network to identify the screw position, establish the screw feature vector, and calculate the group feature center vector and the characteristic difference distance threshold to achieve no manual contact. Quick detection.
It improves the accuracy and efficiency of the loose detection of fan tower lifting screws, reduces the need for manual inspection, reduces the possibility of detection errors, and improves the management level of wind farms.
Smart Images

Figure CN118261884B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wind turbine equipment detection, and in particular to a method for detecting loosening of wind turbine tower hoisting screws based on a feature difference distance threshold. Background Art
[0002] Wind turbine tower mounting screws are the key factor in maintaining the stability of the overall wind turbine frame. If the screws are loose, the wind turbine may break in strong winds, causing huge losses. Therefore, carrying out rapid and effective wind turbine tower mounting screw loosening detection is of great significance for improving inspection efficiency, reducing maintenance costs and improving the level of modern management of wind farms.
[0003] There are many reasons for the loosening of wind turbine tower hoisting screws, such as: the screws become loose due to the vibration of the wind turbine tower; the screw hoisting installer fails to tighten or the operation is not standardized, causing some screws to loosen.
[0004] At present, the typical method used by the industry to detect loose screws in wind turbine tower hoisting is: send a special person to test the screws one by one to find loose screws; if the number of screws is small, then there is no problem with this detection method; however, the wind turbine tower hoisting is connected in layers, and each layer contains more than 70 screws. Even detecting one layer takes a long time; at the same time, on the one hand, a single wind turbine needs to climb nearly 100 meters of tower, and check nearly 1,000 screws one by one at each connection layer, which is a great challenge to the patience, sense of responsibility and physical strength of the inspectors; on the other hand, the types, colors and manufacturers of screws in the process of wind turbine tower hoisting are different. Ordinary inspectors need to have a lot of knowledge to thoroughly grasp the status of the corresponding screws, and grassroots inspectors often find it difficult to reach the corresponding technical level. Therefore, even if the inspection efforts are increased, the inspection efficiency is still very low. At the same time, manual inspection is prone to errors, resulting in low detection accuracy, which can easily lead to wind turbine collapse accidents and cause huge losses. Summary of the invention
[0005] The purpose of the present invention is to solve the problems of low detection efficiency and accuracy in the existing wind turbine tower hanging screw loosening detection method, and proposes a wind turbine tower hanging screw loosening detection method based on feature difference distance threshold.
[0006] The specific process of the wind turbine tower hoisting screw loosening detection method based on feature difference distance threshold is as follows:
[0007] S1. Obtain image data FImg of wind turbine tower hanging screws, and use FImg to obtain basic data set FList of wind turbine tower hanging screws;
[0008] S2. Establish a tower hanging screw feature vector operator OptFRow, where the input of OptFRow is a row OptFRowInput of the wind turbine tower hanging screw basic data set FList, and the output is the motor hanging screw feature output vector OptFRowOutput;
[0009] S3, using the OptFRow obtained in S2 to process each element in FList, and obtain the wind turbine tower hanging screw feature vector corresponding to each element in FList;
[0010] S4, using the wind turbine tower hanging screw feature vector TZVector, obtain the tower hanging screw group feature center vector GroupVector and feature difference distance threshold DifferentGap;
[0011] S5. Use the DifferentGap and GroupVector obtained in S4 to detect the looseness of the wind turbine tower hanging screws and obtain the detection results.
[0012] Furthermore, the image data FImg of the wind turbine tower hanging screws is obtained in S1, and the basic data set FList of the wind turbine tower hanging screws is obtained by using FImg, specifically:
[0013] S101, obtain the image data FImg of the wind turbine tower hanging screws, and use FImg to obtain the identification frame set FSBList:
[0014] Each recognition box contains the following information: the coordinates of the center point of the recognition box (LX, LY), the width of the recognition box LW, and the height of the recognition box LH; LX is the X-axis coordinate of the center point of the recognition box, and LY is the Y-axis coordinate of the center point of the recognition box;
[0015] S102, set the recognition result counter FCounter=1;
[0016] S103, establishing an empty wind turbine tower hanging screw basic data set FList;
[0017] S104, basic data row FRow for the new wind turbine tower hoisting screws;
[0018] S105, set the ID field of FRow to FCounter;
[0019] S106. Get the X-axis position field PosX of FRow:
[0020] PosX = X-axis position of FSBList[FCounter] = LX / FImgW
[0021] Where FImgW is the width of FImg;
[0022] S107. Get the Y-axis position field PosY of FRow:
[0023] PosY = Y-axis position of FSBList[FCounter] = LY / FImgH
[0024] Where FImgH is the height of FImg;
[0025] S108. Get the width field PosW of FRow:
[0026] PosW = FSBList[FCounter] width = LW / FImgW
[0027] S109. Get the height field PosH of FRow:
[0028] PosH = height of FSBList[FCounter] = LH / FImgH
[0029] S110. Get the Y-axis position field PosFSY of the anti-loosening pad of FRow:
[0030] PosFSY = FSBList[FCounter] Description area within the screw lock washer identification frame center point Y axis position = FSY / FImgH
[0031] The screw anti-loosening pad is obtained by using Yolo neural network recognition;
[0032] S111. Get the nut Y-axis position field PosLM of FRow:
[0033] PosLM = FSBList [FCounter] describes the Y-axis position of the nut center point within the area / FImgH
[0034] The FSBList[FCounter] describes the identification of nuts within the area using the Yolo neural network;
[0035] S112. Get the screw Y-axis position field PosLS of FRow:
[0036] PosLS = FSBList[FCounter] describes the Y-axis position of the screw center point within the area / FImgH
[0037] The FSBList[FCounter] describes the identification of screws within the area using the Yolo neural network;
[0038] S113. Get the wind turbine tower hoisting screw feature vector field TZVector of FRow:
[0039] TZVector = vector with 6 elements
[0040] All elements in the said vector have an initial value of 0;
[0041] S114. Add FRow to FList;
[0042] S115. Let FCounter = FCounter + 1;
[0043] S116. If FCounter < the number of elements in FSBList, go to S104; otherwise, output the current FList.
[0044] Furthermore, for obtaining the image data FImg of the hoisting screws of the wind turbine tower in S101 and obtaining the recognition frame set FSBList by using FImg, specifically:
[0045] Use the Yolo neural network to recognize all screw and nut structures in FImg, so as to obtain the recognition frames of the positions of the screw and nut structures in each FImg respectively, and all the recognition frames form the recognition frame set FSBList.
[0046] Furthermore, for establishing the feature vector operator OptFRow of the tower crane hoisting screws in S2, the input of OptFRow is a row OptFRowInput of the basic data set FList of the hoisting screws of the wind turbine tower, and the output is the feature output vector OptFRowOutput of the motor hoisting screws, specifically:
[0047] S201. Establish the feature vector operator OptFRow of the tower crane hoisting screws, and the input of OptFRow is a row OptFRowInput of the basic data set FList of the hoisting screws of the wind turbine tower;
[0048] S202. Obtain the first temporary variable OT1 of the feature vector operator of the tower crane hoisting screws:
[0049]
[0050] Among them, OptFRowInput.PosY is the Y-axis position field in OptFRowInput, OptFRowInput.PosFSY is the anti-loosening pad Y-axis position field in OptFRowInput, OptFRowInput.PosH is the height field in OptFRowInput, and arctan is the arctangent function;
[0051] S203. Obtain the second temporary variable OT2 of the feature vector operator of the tower crane hoisting screws:
[0052]
[0053] S204, obtaining the third temporary variable OT3 of the tower hoisting screw characteristic vector operator:
[0054]
[0055] Among them, OptFRowInput.PosLM is the nut Y-axis position field in OptFRowInput;
[0056] S205, obtaining the fourth temporary variable OT4 of the characteristic vector operator of the wind turbine tower hanging screw:
[0057]
[0058] Among them, OptFRowInput.PosW is the width field in OptFRowInput;
[0059] S206, obtaining the fifth temporary variable OT5 of the characteristic vector operator of the wind turbine tower hanging screw:
[0060]
[0061] Among them, Exp is the exponential function with the natural constant e as the base, and OptFRowInput.PosX is the X-axis field in OptFRowInput;
[0062] S207, obtaining the sixth temporary variable OT6 of the characteristic vector operator of the wind turbine tower hanging screw:
[0063]
[0064] S208, using the first temporary variable of the wind turbine tower hanging screw characteristic vector operator obtained in S202-S207 to the sixth temporary variable of the wind turbine tower hanging screw characteristic vector operator to obtain the wind turbine tower hanging screw characteristic output vector OptFRowOutput = (OT1, OT2, OT3, OT4, OT5, OT6);
[0065] S209. Return OptFRowOutput as OptFRow.
[0066] Furthermore, the OptFRow obtained in S2 is used in S3 to process each element in FList to obtain the wind turbine tower hanging screw feature vector corresponding to each element in FList, specifically:
[0067] S301, set the feature calculation counter TZCounter=1;
[0068] S302. Let the feature calculation temporary row variable TZRow be the TZCounter-th element of FList:
[0069] S303. Obtain the feature calculation temporary result TZTemp:
[0070] The feature calculation temporary result TZTemp = perform calculation using OptFRow, where OptFRowInput = TZRow;
[0071] The wind turbine tower crane lifting screw feature vector TZVector of the TZCounter-th element of FList = TZTemp;
[0072] S305. Let TZCounter = TZCounter + 1;
[0073] S306. If TZCounter < the number of elements in FList, go to S302; otherwise, output the wind turbine tower crane lifting screw feature vectors corresponding to each element of FList.
[0074] Furthermore, in S4, using the wind turbine tower crane lifting screw feature vector TZVector to obtain the tower crane lifting screw group feature center vector GroupVector and the feature difference distance threshold DifferentGap is specifically as follows:
[0075] S401. Let the wind turbine tower crane lifting screw group feature center vector GroupVector = the mean of the TZVector fields of all elements in FList;
[0076] S402. Use the KMeans algorithm to perform clustering calculation on the TZVector field of FList with 2 categories, and obtain the first cluster Cluster1 and the second cluster Cluster2 respectively;
[0077] S403. The number of elements in the first cluster Cluster1Num = obtain the number of internal elements in Cluster1;
[0078] S404. The number of elements in the second cluster Cluster2Num = obtain the number of internal elements in Cluster2;
[0079] S405. The center of the first cluster Cluster1Center = the center point vector of Cluster1;
[0080] S406. The center of the second cluster Cluster2Center = the center point vector of Cluster2;
[0081] S407, dual center span variable Kuadu = l2norm (Cluster1Center - Cluster2Center) × 0.55;
[0082] Among them, l2norm is the Euclidean norm;
[0083] S408, if Cluster1Num is greater than Cluster2Num, go to S409, otherwise go to S411;
[0084] S409, GroupVector=(GroupVector+Cluster1Center) / 2;
[0085] S410, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster1Center)+Kuadu, go to S413;
[0086] S411. GroupVector=(GroupVector+Cluster2Center) / 2;
[0087] S412, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster2Center)+Kuadu, and go to S414;
[0088] S413, output the feature difference distance threshold DifferentGap and the feature center vector GroupVector of the current wind turbine tower hoisting screw group.
[0089] Furthermore, the DifferentGap and GroupVector obtained in S4 are used in S5 to detect the looseness of the wind turbine tower hanging screws to obtain the detection results, which are specifically:
[0090] S501, set the detection counter TestCounter=1;
[0091] S502, let the detection calculation temporary row variable TestRow = the TestCounterth element of FList;
[0092] S503, detect and calculate the temporary distance variable TestTempDistance = l2norm (TestRow.TZVector-GroupVector);
[0093] Among them, TestRow.TZVector is the TZVector attribute of TestRow;
[0094] S504. If TestTempDistance is less than or equal to DifferentGap, go to S506; otherwise, go to S505.
[0095] S505. The screw corresponding to TestRow is loose. Output the loose screw at the position where the X coordinate in FImg is TestRow.PosX and the Y coordinate is TestRow.PosY.
[0096] S506. TestCounter = TestCounter + 1.
[0097] S507. If TestCounter < the number of elements in FList, go to S502; otherwise, go to S508.
[0098] S508. The whole process ends and the detection result is output.
[0099] The beneficial effects of the present invention are as follows:
[0100] The present invention provides a method for detecting the looseness of hoisting screws of a wind turbine tower based on a feature difference distance threshold. First, the visual features of all hoisting screws in the image data of the hoisting screws of the wind turbine tower are described, and then the feature difference distance threshold is constructed. According to the feature, whether the distance between different screws and the whole screw group exceeds the feature difference distance threshold is used to detect the looseness of the hoisting screws of the wind turbine tower. According to the law that most of the screws on each layer of the hoisting tower of the wind turbine tower are fastened and only a small part are loose, the present invention can identify the loose screws without manual contact and prior knowledge, and realizes fast and effective detection of the looseness of the hoisting screws. The present invention does not require manual inspection, and improves the accuracy and efficiency of detecting the looseness of the hoisting screws of the wind turbine tower. BRIEF DESCRIPTION OF THE DRAWINGS
[0101] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0102] DETAILED DESCRIPTION OF THE INVENTION I: As Figure 1 shown, the specific process of the method for detecting the looseness of hoisting screws of a wind turbine tower based on a feature difference distance threshold in this embodiment is as follows:
[0103] S1. Obtain the image data FImg of the hoisting screws of the wind turbine tower through a camera and an internal camera of the wind turbine, and use FImg to obtain the basic data set FList of the hoisting screws of the wind turbine tower. Specifically:
[0104] S101. Obtain the image data FImg of the hoisting screws of the wind turbine tower through a camera and an internal camera of the wind turbine, and use FImg to obtain the recognition frame set FSBList:
[0105] The Yolo neural network is used to identify all screw and nut structures in FImg, so as to obtain the identification boxes of the screw and nut structure positions in each FImg respectively. All the identification boxes form the identification box set FSBList;
[0106] Each recognition box contains the following information: the coordinates of the center point of the recognition box (LX, LY), the width of the recognition box LW, and the height of the recognition box LH; LX is the X-axis coordinate of the center point of the recognition box, and LY is the Y-axis coordinate of the center point of the recognition box;
[0107] S102, set the recognition result counter FCounter=1;
[0108] S103, establishing an empty wind turbine tower hanging screw basic data set FList;
[0109] S104, basic data row FRow for the new wind turbine tower hoisting screws;
[0110] S105, set the ID field of FRow to FCounter;
[0111] S106. Get the X-axis position field PosX of FRow:
[0112] PosX = X-axis position of FSBList[FCounter] = LX / FImgW
[0113] Where FImgW is the width of FImg;
[0114] S107. Get the Y-axis position field PosY of FRow:
[0115] PosY = Y-axis position of FSBList[FCounter] = LY / FImgH
[0116] Where FImgH is the height of FImg;
[0117] S108. Get the width field PosW of FRow:
[0118] PosW = FSBList[FCounter] width = LW / FImgW
[0119] S109. Get the height field PosH of FRow:
[0120] PosH = height of FSBList[FCounter] = LH / FImgH
[0121] S110. Get the Y-axis position field PosFSY of the anti-loosening pad of FRow:
[0122] PosFSY = The Y-axis position of the center point of the anti-loosening pad recognition frame for screws within the described area = FSY / FImgH
[0123] The anti-loosening pad for screws is obtained by recognition using the Yolo neural network;
[0124] S111. Obtain the Y-axis position field PosLM of the nut in FRow:
[0125] PosLM = The Y-axis position of the center point of the nut within the described area of FSBList[FCounter] / FImgH
[0126] The nut within the described area of FSBList[FCounter] is recognized using the Yolo neural network;
[0127] S112. Obtain the Y-axis position field PosLS of the screw in FRow:
[0128] PosLS = The Y-axis position of the center point of the screw within the described area of FSBList[FCounter] / FImgH
[0129] The screw within the described area of FSBList[FCounter] is recognized using the Yolo neural network;
[0130] S113. Obtain the feature vector field TZVector of the wind turbine tower barrel hoisting screw in FRow:
[0131] TZVector = A vector with 6 elements
[0132] All elements in the said vector have an initial value of 0;
[0133] S114. Add FRow to FList;
[0134] S115. Let FCounter = FCounter + 1;
[0135] S116. If FCounter < the number of elements in FSBList, then go to S104, otherwise output the current FList.
[0136] S2. Establish the feature vector operator OptFRow for the tower barrel hoisting screw. The input of OptFRow is a row OptFRowInput of the basic data set FList of the wind turbine tower barrel hoisting screw, and the output is the feature output vector OptFRowOutput of the motor hoisting screw. Specifically:
[0137] S201, establish a tower hanging screw feature vector operator OptFRow, where the OptFRow input is a row OptFRowInput of the wind turbine tower hanging screw basic data set FList;
[0138] S202, obtain the first temporary variable OT1 of the characteristic vector operator of the wind turbine tower hanging screw:
[0139]
[0140] Among them, OptFRowInput.PosY is the Y-axis position field in OptFRowInput, OptFRowInput.PosFSY is the Y-axis position field of the anti-loosening pad in OptFRowInput, OptFRowInput.PosH is the height field in OptFRowInput, arctan inverse tangent function;
[0141] S203, obtaining the second temporary variable OT2 of the characteristic vector operator of the wind turbine tower hanging screw:
[0142]
[0143] S204, obtaining the third temporary variable OT3 of the tower hoisting screw characteristic vector operator:
[0144]
[0145] Among them, OptFRowInput.PosLM is the nut Y-axis position field in OptFRowInput;
[0146] S205, obtaining the fourth temporary variable OT4 of the characteristic vector operator of the wind turbine tower hanging screw:
[0147]
[0148] Among them, OptFRowInput.PosW is the width field in OptFRowInput;
[0149] S206, obtaining the fifth temporary variable OT5 of the characteristic vector operator of the wind turbine tower hanging screw:
[0150]
[0151] Among them, Exp is the exponential function with the natural constant e as the base, and OptFRowInput.PosX is the X-axis field in OptFRowInput;
[0152] S207, obtaining the sixth temporary variable OT6 of the characteristic vector operator of the wind turbine tower hanging screw:
[0153]
[0154] S208. Obtain the first temporary variable to the sixth temporary variable of the feature vector operator of the wind turbine tower crane lifting screw using the ones obtained in S202 - S207, and obtain the feature output vector of the wind turbine tower crane lifting screw OptFRowOutput = (OT1, OT2, OT3, OT4, OT5, OT6);
[0155] S209. Return OptFRowOutput as OptFRow.
[0156] S3. Process each element in FList using OptFRow obtained in S2 to obtain the feature vector of the wind turbine tower crane lifting screw corresponding to each element in FList, specifically:
[0157] S301. Let the feature calculation counter TZCounter = 1;
[0158] S302. Let the feature calculation temporary row variable TZRow be the TZCounter-th element of FList:
[0159] S303. Obtain the feature calculation temporary result TZTemp:
[0160] The feature calculation temporary result TZTemp = perform calculation using OptFRow, OptFRowInput = TZRow;
[0161] The feature vector TZVector of the wind turbine tower crane lifting screw of the TZCounter-th element in FList = TZTemp;
[0162] S305. Let TZCounter = TZCounter + 1;
[0163] S306. If TZCounter < the number of elements in FList, go to S302; otherwise, output the feature vector of the wind turbine tower crane lifting screw corresponding to each element in FList.
[0164] S4. Use the feature vector TZVector of the wind turbine tower crane lifting screw in FList to obtain the group feature center vector GroupVector and the feature difference distance threshold DifferentGap of the tower crane lifting screw group, specifically:
[0165] S401. Let the group feature center vector GroupVector of the wind turbine tower crane lifting screw be the mean of the TZVector fields of all elements in FList;
[0166] S402, using the KMeans algorithm to perform a clustering calculation of the TZVector field of FList with the category of 2, and obtain the first cluster Cluster1 and the second cluster Cluster2 respectively;
[0167] S403, the number of elements in the first cluster Cluster1Num=the number of elements in Cluster1 is obtained;
[0168] S404, the number of elements of the second cluster Cluster2Num=the number of elements inside Cluster2 is obtained;
[0169] S405, the center of the first cluster Cluster1Center=the center point vector of Cluster1;
[0170] S406, the center of the second cluster Cluster2Center=the center point vector of Cluster2;
[0171] S407, dual center span variable Kuadu = l2norm (Cluster1Center - Cluster2Center) × 0.55;
[0172] Among them, l2norm is the Euclidean norm;
[0173] S408, if Cluster1Num is greater than Cluster2Num, go to S409, otherwise go to S411;
[0174] S409, GroupVector=(GroupVector+Cluster1Center) / 2;
[0175] S410, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster1Center)+Kuadu, go to S413;
[0176] S411. GroupVector=(GroupVector+Cluster2Center) / 2;
[0177] S412, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster2Center)+Kuadu, and go to S414;
[0178] S413, output the feature difference distance threshold DifferentGap and the feature center vector GroupVector of the current wind turbine tower hoisting screw group.
[0179] S5. Use the DifferentGap and GroupVector obtained in S4 to detect the looseness of the hoisting screws of the wind turbine tower, and obtain the detection result, specifically:
[0180] S501. Let the detection counter TestCounter = 1;
[0181] S502. Let the detection calculation temporary row variable TestRow be the TestCounter-th element of FList;
[0182] S503. The detection calculation temporary distance variable TestTempDistance = l2norm(TestRow.TZVector - GroupVector);
[0183] where TestRow.TZVector is the TZVector attribute of TestRow;
[0184] S504. If TestTempDistance is less than or equal to DifferentGap, then go to S506; otherwise, go to S505;
[0185] S505. The screw corresponding to TestRow is loose, and output that there is a loose screw at the position where the X coordinate in FImg is TestRow.PosX and the Y coordinate is TestRow.PosY;
[0186] S506. TestCounter = TestCounter + 1;
[0187] S507. If TestCounter < the number of elements in FList, then go to S502; otherwise, go to S508;
[0188] S508. The whole process ends, and output the detection result.
Claims
1. A wind turbine tower hoisting screw loosening detection method based on feature difference distance threshold, characterized by The specific process of the method is: S1. Obtain image data FImg of wind turbine tower hanging screws, and use FImg to obtain basic data set FList of wind turbine tower hanging screws; S2. Establish a tower hanging screw feature vector operator OptFRow, where the input of OptFRow is a row OptFRowInput of the wind turbine tower hanging screw basic data set FList, and the output is the motor hanging screw feature output vector OptFRowOutput; S3, using the OptFRow obtained in S2 to process each element in FList, and obtain the wind turbine tower hanging screw feature vector corresponding to each element in FList; S4. Using the wind turbine tower hanging screw feature vector TZVector, obtain the tower hanging screw group feature center vector GroupVector and feature difference distance threshold DifferentGap, specifically: S401, let the wind turbine tower hanging screw group feature center vector GroupVector = the mean value of the TZVector field of all elements in FList; S402, using the KMeans algorithm to perform a clustering calculation of the TZVector field of FList with the category of 2, and obtain the first cluster Cluster1 and the second cluster Cluster2 respectively; S403, the number of elements in the first cluster Cluster1Num=the number of elements in Cluster1 is obtained; S404, the number of elements of the second cluster Cluster2Num=the number of elements inside Cluster2 is obtained; S405, the center of the first cluster Cluster1Center=the center point vector of Cluster1; S406, the center of the second cluster Cluster2Center=the center point vector of Cluster2; S407, dual center span variable Kuadu = l2norm (Cluster1Center - Cluster2Center) × 0.55; Among them, l2norm is the Euclidean norm; S408, if Cluster1Num is greater than Cluster2Num, go to S409, otherwise go to S411; S409, GroupVector=(GroupVector+Cluster1Center) / 2; S410, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster1Center)+Kuadu, go to S413; S411. GroupVector=(GroupVector+Cluster2Center) / 2; S412, feature difference distance threshold DifferentGap=l2norm(GroupVector-Cluster2Center)+Kuadu, and go to S414; S413, outputting a feature difference distance threshold DifferentGap and a feature center vector GroupVector of the current wind turbine tower hoisting screw group; S5. Use the DifferentGap and GroupVector obtained in S4 to detect the looseness of the wind turbine tower hanging screws and obtain the detection results.
2. The method for detecting wind turbine tower hanging screw loosening based on feature difference distance threshold according to claim 1 is characterized in that: The image data FImg of the wind turbine tower hanging screws is obtained in S1, and the basic data set FList of the wind turbine tower hanging screws is obtained by using FImg, specifically: S101, obtain the image data FImg of the wind turbine tower hanging screws, and use FImg to obtain the identification frame set FSBList: Each recognition box contains the following information: the coordinates of the center point of the recognition box (LX, LY), the width of the recognition box LW, and the height of the recognition box LH; LX is the X-axis coordinate of the center point of the recognition box, and LY is the Y-axis coordinate of the center point of the recognition box; S102, set the recognition result counter FCounter=1; S103, establishing an empty wind turbine tower hanging screw basic data set FList; S104, basic data row FRow for the new wind turbine tower hoisting screws; S105, set the ID field of FRow to FCounter; S106. Get the X-axis position field PosX of FRow: PosX = X-axis position of FSBList[FCounter] = LX / FImgW Where FImgW is the width of FImg; S107. Get the Y-axis position field PosY of FRow: PosY = Y-axis position of FSBList[FCounter] = LY / FImgH Where FImgH is the height of FImg; S108. Get the width field PosW of FRow: PosW = FSBList[FCounter] width = LW / FImgW S109. Get the height field PosH of FRow: PosH = height of FSBList[FCounter] = LH / FImgH S110. Get the Y-axis position field PosFSY of the anti-loosening pad of FRow: PosFSY = FSBList[FCounter] Description area within the screw lock washer identification frame center point Y axis position = FSY / FImgH The screw anti-loosening pad is obtained by using Yolo neural network recognition; S111. Get the nut Y-axis position field PosLM of FRow: PosLM = FSBList [FCounter] describes the Y-axis position of the nut center point within the area / FImgH The FSBList[FCounter] describes the identification of nuts within the area using the Yolo neural network; S112. Get the screw Y-axis position field PosLS of FRow: PosLS = FSBList[FCounter] describes the Y-axis position of the screw center point within the area / FImgH The FSBList[FCounter] describes the identification of screws within the area using the Yolo neural network; S113. Obtain the wind turbine tower crane lifting screw feature vector field TZVector of FRow: TZVector = vector of 6 elements All elements in the vector have an initial value of 0; S114. Add FRow to FList; S115. Let FCounter = FCounter + 1; S116. If FCounter < the number of elements in FSBList, go to S104, otherwise output the current FList.
3. The method for detecting wind turbine tower hanging screw loosening based on feature difference distance threshold according to claim 2 is characterized in that: The acquisition of the image data FImg of the wind turbine tower crane lifting screw in S101 and the acquisition of the recognition frame set FSBList using FImg are specifically as follows: Use the Yolo neural network to recognize all screw and nut structures in FImg, so as to obtain the recognition frames of the positions of the screw and nut structures in each FImg respectively. All the recognition frames form the recognition frame set FSBList.
4. The method for detecting wind turbine tower hanging screw loosening based on feature difference distance threshold according to claim 3 is characterized in that: The establishment of the tower crane lifting screw feature vector operator OptFRow in S2, where OptFRow takes a row OptFRowInput of the basic data set FList of the wind turbine tower crane lifting screw as the input and outputs the motor lifting screw feature output vector OptFRowOutput, is specifically as follows: S201. Establish the tower crane lifting screw feature vector operator OptFRow, and OptFRow takes a row OptFRowInput of the basic data set FList of the wind turbine tower crane lifting screw as the input; S202. Obtain the first temporary variable OT1 of the wind turbine tower crane lifting screw feature vector operator: Among them, OptFRowInput.PosY is the Y-axis position field in OptFRowInput, OptFRowInput.PosFSY is the anti-loosening pad Y-axis position field in OptFRowInput, OptFRowInput.PosH is the height field in OptFRowInput, and arctan is the arctangent function; S203. Obtain the second temporary variable OT2 of the wind turbine tower crane lifting screw feature vector operator: S204. Obtain the third temporary variable OT3 of the tower crane lifting screw feature vector operator: Among them, OptFRowInput.PosLM is the nut Y-axis position field in OptFRowInput; S205. Obtain the fourth temporary variable OT4 of the wind turbine tower crane lifting screw feature vector operator: Among them, OptFRowInput.PosW is the width field in OptFRowInput; S206. Obtain the fifth temporary variable OT5 of the wind turbine tower crane lifting screw feature vector operator: Among them, Exp is the exponential function with the natural constant e as the base, and OptFRowInput.PosX is the X-axis field in OptFRowInput; S207. Obtain the sixth temporary variable OT6 of the wind turbine tower crane lifting screw feature vector operator: S208. Obtain the characteristic output vector of the wind turbine tower crane lifting screw OptFRowOutput = (OT1, OT2, OT3, OT4, OT5, OT6) from the first temporary variable to the sixth temporary variable of the characteristic vector operator of the wind turbine tower crane lifting screw obtained by S202 - S207; S209. Return OptFRowOutput as OptFRow.
5. The method for detecting wind turbine tower hanging screw loosening based on feature difference distance threshold according to claim 4 is characterized in that: In step S3, process each element in FList using OptFRow obtained in S2 to obtain the characteristic vector of the wind turbine tower crane lifting screw corresponding to each element in FList. Specifically: S301. Let the characteristic calculation counter TZCounter = 1; S302. Let the temporary row variable for characteristic calculation TZRow be the TZCounter-th element of FList: S303. Obtain the temporary result of characteristic calculation TZTemp: The temporary result of characteristic calculation TZTemp = perform calculation using OptFRow, where OptFRowInput = TZRow; S304. The characteristic vector TZVector of the wind turbine tower crane lifting screw for the TZCounter-th element of FList = TZTemp; S305. Let TZCounter = TZCounter + 1; S306. If TZCounter < the number of elements in FList, go to S302; otherwise, output the characteristic vectors of the wind turbine tower crane lifting screws corresponding to each element in FList.
6. The method for detecting wind turbine tower hanging screw loosening based on feature difference distance threshold according to claim 5 is characterized in that: In step S5, detect the loosening condition of the wind turbine tower crane lifting screw using DifferentGap and GroupVector obtained in S4 to obtain the detection result. Specifically: S501. Let the detection counter TestCounter = 1; S502. Let the temporary row variable for detection calculation TestRow be the TestCounter-th element of FList; S503. The temporary distance variable for detection calculation TestTempDistance = l2norm(TestRow.TZVector - GroupVector); Here, TestRow.TZVector is the TZVector property of TestRow; S504. If TestTempDistance is less than or equal to DifferentGap, go to S506; otherwise, go to S505; S505. The screw corresponding to TestRow is loose, and output that there is a loose screw at the position where the X coordinate in FImg is TestRow.PosX and the Y coordinate is TestRow.PosY; S506. Let TestCounter = TestCounter + 1; S507. If TestCounter < the number of elements in FList, go to S502; otherwise, go to S508; S508. The whole process ends, and output the detection result.
